Two-stage CNN model training method and system, damage area identification method and system

Abstract

The present invention belongs to the field of carbon fiber composite material structural health monitoring and provides a two-stage CNN model training method and system, and a damage area identification method and system. The training method includes constructing a measured simulated damage sample data set and a numerically calculated damage sample data set; constructing a structural damage sample database based on the measured simulated damage sample data set and the numerically calculated damage sample data set; using the structural damage sample database, respectively training a first CNN model and a second CNN model to obtain a first characteristic matrix and a second characteristic matrix, respectively; and training a third CNN model based on a fusion matrix of the first characteristic matrix and the second characteristic matrix. In actual application, the normalized damage scattering signal envelope collected by the sensor array is input into the trained two-stage CNN network prediction model to achieve rapid and accurate identification of the damaged area of the carbon fiber composite material structure.

Description

Technical Field

[0001] The present invention belongs to the field of carbon fiber composite material structure health monitoring, and in particular relates to a two-stage CNN model training method and system, and a damage area identification method and system. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Carbon fiber composites, with their excellent mechanical properties, such as high specific strength, high specific modulus, and corrosion resistance, are widely used in aerospace, rail transportation, and other fields. They are currently popular non-metallic composite materials that can replace metal materials as structural materials. Although carbon fiber composites have incomparable advantages in mechanical and chemical properties, during service in complex and changing external environments, material properties may degrade, resulting in subtle damage such as fatigue cracks and delamination that are difficult to detect. When the damage expands to a certain extent, it may cause overall structural failure, threatening the safe operation of equipment.

[0004] Lamb waves have been widely used in the field of structural health monitoring due to their advantages of low detection cost, wide monitoring area, high sensitivity, and non-destructive nature. Their principle is to use sensors coupled to or embedded within the structure to collect structural state signals. Advanced techniques are then used to extract damage-related signal features to determine the damage location. With the development of intelligent algorithms, deep learning-based structural damage detection has become a research hotspot. Numerous researchers have proposed structural damage detection methods combining various deep learning models with Lamb wave technology and conducted experiments to verify the feasibility and accuracy of the proposed methods. Although deep learning-based Lamb wave damage detection technology provides a new approach for interpreting and fully mining complex guided wave signals in anisotropic carbon fiber composite structures and exhibits excellent structural generalization capabilities, this method requires a certain number of training samples in advance, and the model structure and parameters are often determined through trial-and-error methods. Currently, a unified and effective method for determining model parameters has not yet been established. Furthermore, the characteristic differences between test and training samples are a significant factor that hinders the practical application of intelligent damage detection technology. Summary of the Invention

[0005] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a two-stage CNN model training method and system, and a damage area identification method and system. By utilizing CNN networks of different sizes and structures to establish a two-stage damage area identification model, it can not only effectively reduce the extremely high requirements of large-area monitoring areas on the regional resolution capability of deep learning network models, but also reduce the excessive dependence of model performance on parameters, thereby improving the accuracy of damage area identification of carbon fiber composite materials.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] A first aspect of the present invention provides a two-stage CNN model training method.

[0008] A two-stage CNN model training method, comprising:

[0009] Construct a measured simulated damage sample data set and a numerically calculated damage sample data set; construct a structural damage sample database based on the measured simulated damage sample data set and the numerically calculated damage sample data set;

[0010] The structural damage sample database is used to train the first CNN model and the second CNN model respectively to obtain the first feature matrix and the second feature matrix respectively; the third CNN model is trained based on the fusion matrix of the first feature matrix and the second feature matrix.

[0011] Furthermore, the process of constructing the measured simulated damage sample data set includes:

[0012] Obtain the guided wave signal when the structure is in a healthy state and mark it as a healthy signal;

[0013] Simulate structural damage and obtain the guided wave signal after the structure is damaged, which is marked as the damage signal;

[0014] According to the healthy signal and the damaged signal, the scattered signal corresponding to the damaged signal is obtained;

[0015] The envelope of the scattered signal is calculated using Hilbert transform, and a dataset of measured simulated damage samples of the scattered signal envelope is constructed.

[0016] Furthermore, the process of constructing the numerical calculation damage sample data set includes:

[0017] Based on the guided wave signal in the healthy state of the structure, the propagation velocity of Lamb waves in each direction of the carbon fiber composite material structure is fitted;

[0018] Based on the pre-fitted propagation velocity of Lamb waves in all directions of the carbon fiber composite material structure, a numerical calculation damage sample data set is constructed.

[0019] A second aspect of the present invention provides a dual-stage CNN model training system.

[0020] A two-stage CNN model training system, comprising:

[0021] A database construction module is configured to: construct a measured simulated damage sample data set and a numerically calculated damage sample data set; and construct a structural damage sample database based on the measured simulated damage sample data set and the numerically calculated damage sample data set;

[0022] The model training module is configured to: use a structural damage sample database to train the first CNN model and the second CNN model respectively, to obtain the first feature matrix and the second feature matrix respectively; and train the third CNN model based on the fusion matrix of the first feature matrix and the second feature matrix.

[0023] A third aspect of the present invention provides a dual-stage CNN model training system.

[0024] A two-stage CNN model training system includes: a piezoelectric sensor array, an ultrasonic guided wave monitoring instrument, and a host computer, wherein the piezoelectric sensor array is coupled to a carbon fiber composite material structure, the piezoelectric sensor array is connected to the ultrasonic guided wave monitoring instrument, and the ultrasonic guided wave monitoring instrument is connected to the host computer; wherein,

[0025] The piezoelectric sensor array is used to collect guided wave signals when the structure is in a healthy state and guided wave signals after the structure is damaged, and transmit them to the host computer through the ultrasonic guided wave monitoring instrument;

[0026] The host computer obtains a scattered signal envelope based on the guided wave signal in a healthy state of the structure and the guided wave signal after the structure is damaged, and constructs a measured simulated damage sample data set of the scattered signal envelope; and fits the propagation speed of Lamb waves in various directions of the carbon fiber composite material structure based on the guided wave signal in a healthy state of the structure to construct a numerical calculation damage sample data set; constructs a structural damage sample database based on the measured simulated damage sample data set and the numerical calculation damage sample data set; uses the structural damage sample database to train a first CNN model and a second CNN model, respectively, to obtain a first feature matrix and a second feature matrix, respectively; and trains a third CNN model based on a fusion matrix of the first feature matrix and the second feature matrix.

[0027] A fourth aspect of the present invention provides a damaged area identification method.

[0028] A method for identifying a damaged area, comprising:

[0029] Based on the scattered signal of the damage to be detected, the signals are input into the first CNN model and the second CNN model respectively to obtain a first prediction result matrix and a second prediction result matrix respectively;

[0030] Obtaining a preliminary prediction result matrix based on the product of the first prediction result matrix and the second prediction result matrix;

[0031] The third CNN model is used to predict the preliminary prediction result matrix to obtain the regional prediction result of the damage to be tested;

[0032] Among them, the first CNN model, the second CNN model and the third CNN model are all obtained through the two-stage CNN model training method described in the first aspect.

[0033] Furthermore, obtaining the preliminary prediction result matrix includes:

[0034] If the preliminary prediction result matrix is an n×m-dimensional preliminary prediction result matrix;

[0035] Then the element at position (g, j) in the matrix is the product of the number of occurrences of the region ((g-1)×m+j) in the prediction result of the first CNN model and the prediction result of the second CNN model.

[0036] A fifth aspect of the present invention provides a damaged area identification system.

[0037] A damaged area identification system, comprising:

[0038] A primary prediction module is configured to: input the scattered signals of the damage to be detected into the first CNN model and the second CNN model respectively, to obtain a first prediction result matrix and a second prediction result matrix respectively;

[0039] A fusion module is configured to: obtain a preliminary prediction result matrix based on the product of the first prediction result matrix and the second prediction result matrix;

[0040] The secondary prediction module is configured to: use the third CNN model to predict the preliminary prediction result matrix to obtain the regional prediction result of the damage to be detected;

[0041] Among them, the first CNN model, the second CNN model and the third CNN model are all obtained through the two-stage CNN model training method described in the first aspect.

[0042] A sixth aspect of the present invention provides a computer-readable storage medium.

[0043] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the two-stage CNN model training method as described in the first aspect, or implements the steps in the damage area identification method as described in the fourth aspect.

[0044] A seventh aspect of the present invention provides a computer device.

[0045] A computer device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the processor implements the steps of the two-stage CNN model training method described in the first aspect, or implements the steps of the damage area identification method described in the fourth aspect.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] (1) The present invention calculates the arrival time of the direct wave at each discrete point in the monitoring area on the sensing path based on the fitted propagation velocity, and then obtains the normalized damage scattering signal envelope based on the numerical calculation of the time and the excitation signal envelope. This can quickly and inexpensively obtain guided wave data under any number of structural damage states.

[0048] (2) The present invention uses the normalized scattering signal envelope of numerically calculated damage and the normalized scattering signal envelope of measured simulated damage to construct a structural damage sample database, which can not only effectively meet the CNN damage model's demand for massive data driving, but also, due to the difference between the numerically calculated data and the measured data, it is also conducive to ensuring the stability of the damage area identification model.

[0049] (3) The present invention constructs a two-level damage area recognition model based on multiple CNN networks. By performing secondary discrimination on the initially predicted damage area results, it can not only effectively reduce the requirements of multiple categories for the high-resolution capability of the model, but also reduce the dependence of the model discrimination results on the model parameters, thereby improving the accuracy of damage area recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0051] FIG1(a) is a schematic diagram of the monitoring area division of the normalized envelope of the scattered signals of the damage discrete points within the monitoring area shown in the present invention;

[0052] FIG1( b ) is a schematic diagram of a sensing path i during the numerical calculation of the normalized envelope of the scattered signal of a damaged discrete point within the monitoring area shown in the present invention;

[0053] FIG1(c) is a schematic diagram of the numerical calculation of the scattered signal and envelope during the numerical calculation of the normalized envelope of the scattered signal of the damaged discrete points within the monitoring area shown in the present invention;

[0054] FIG1(d) is a numerical calculation diagram of the damage discrete points in the sub-region k during the numerical calculation of the normalized envelope of the scattered signals of the damage discrete points in the monitoring area shown in the present invention;

[0055] FIG1(e) is a diagram showing the normalized envelope of scattered signals of discrete damage points within the monitoring area of the present invention, showing the actual measured simulated damage discrete points within the sub-area k during the damage measurement process;

[0056] FIG2( a ) is a schematic diagram of the regional set division of the CNN1 model set shown in the present invention;

[0057] FIG2( b ) is a schematic diagram of the regional set division of the CNN2 model set shown in the present invention;

[0058] Figure 3 1 is a schematic diagram of a structural damage area recognition model training based on a two-stage CNN network shown in the present invention;

[0059] Figure 4 1 is a schematic diagram of a structural damage area recognition model based on a two-stage CNN network shown in the present invention. DETAILED DESCRIPTION

[0060] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0061] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0062] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0063] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the methods and systems according to the various embodiments of the present disclosure. It should be noted that each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the module, program segment, or a part of code can include one or more executable instructions for implementing the logical functions specified in the various embodiments. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, or they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the flowchart and / or block diagram, and the combination of the boxes in the flowchart and / or block diagram, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or can be implemented using a combination of dedicated hardware and computer instructions.

[0064] Example 1

[0065] like Figure 3 As shown, this embodiment provides a two-stage CNN model training method.

[0066] A two-stage CNN model training method, comprising:

[0067] Construct a measured simulated damage sample data set and a numerically calculated damage sample data set; construct a structural damage sample database based on the measured simulated damage sample data set and the numerically calculated damage sample data set;

[0068] The structural damage sample database is used to train the first CNN model and the second CNN model respectively to obtain the first feature matrix and the second feature matrix respectively; the third CNN model is trained based on the fusion matrix of the first feature matrix and the second feature matrix.

[0069] As shown in Figure 2(a), based on the CNN network, a primary damage area recognition model set CNN1 containing (n-1)×(m-1) CNN models is established: ① All sub-monitoring areas are divided into (n-1)×(m-1) different area sets in sequence according to the size of 2×2 and the step size of 1. For example, sub-areas 1, 2, m+1, and m+2 are the first area set (denoted as area set 1). 1 ), sub-regions 2, 3, m+2, and m+3 are the second region set (denoted as region set 1 2 ), and so on..., the sub-regions (n-1)m-1, (n-1)m, nm-1, nm constitute the region set 1 (n-1)(m-1); ② Randomly generate training sets and validation sets based on the ratio of 8:2 using the damage data samples in each regional set; ③ Use the normalized scattered signal envelope in the training set (data size: T×L) as the input of the CNN network model, and the sub-region corresponding to the normalized envelope (data size: 4×1) as the output of the CNN network model. The number of iterations is 100 to complete the network model parameter training (model structure and parameters are shown in Table 1); ④ Use the normalized scattered signal envelope in the validation set as the model input, compare the model output with the data label (ideal output) in the validation set, and verify the model parameters; ⑤ Repeat steps ③-④ for all regional sets to obtain the primary damage area recognition model set 1 (including CNN1 1 ,CNN1 2 ,…,CNN1 (n -1)(m-1) ).

[0070] Table 1 Parameters of the primary CNN model set

[0071]

[0072] As shown in Figure 2(b), based on the CNN network, a primary damage area recognition model set CNN2 containing (n-2)×(m-2) CNN models is constructed: ① All sub-monitoring areas are divided into (n-2)×(m-2) different area sets in sequence according to the size of 3×3 and the step size of 1. For example, sub-areas 1, 2, 3, m+1, m+2, m+3, 2m+1, 2m+2, 2m+3 are the first area set (denoted as area set 2). 1 ), sub-regions 2, 3, 4, m+2, m+3, m+4, 2m+2, 2m+3, 2m+4 are the second region set (denoted as region set 2 2 ), and so on..., the sub-regions (n-2)m-2, (n-2)m-1, (n-2)m, (n-1)m-2, (n-1)m-1, (n-1)m, nm-2, nm-1, nm are region set 2 (n-2)(m-2) ; ② Randomly generate training sets and validation sets based on the ratio of 8:2 from the damage data samples in each regional set; ③ The normalized scattered signal envelope in the training set (data size: T×L) is used as the input of the CNN network model, and the sub-region corresponding to the normalized scattered signal envelope (data size: 9×1) is used as the output of the CNN network model. The number of iterations is 100 to complete the network model parameter training (model structure and parameters are shown in Table 1); ④ The normalized scattered signal envelope in the validation set is used as the model input, and the model output is compared with the data label (ideal output) in the validation set to verify the model parameters; ⑤ Repeat steps ③-④ for all regional sets to obtain the primary damage area recognition model set 2 (including CNN2 1 ,CNN2 2 ,…,CNN2(n-2)(m-2) ).

[0073] Based on the CNN network, a secondary damage region recognition model CNN3 was constructed: ① The normalized scattered signal envelope (data size: T × L) in the damage sample library was input into the model groups CNN1 and CNN2 respectively. The obtained prediction results were then statistically analyzed to generate an n × m matrix. The element at position (g, j) in the matrix was the product of the CNN1 prediction result and the number of occurrences of the region (g-1) × m + j in the CNN2 prediction result. ② The n × m matrix corresponding to all damage sample libraries was randomly divided into a training set and a validation set in an 8:2 ratio. ③ The n × m matrix in the training set was used as the input of the CNN3 model, and the damage sub-region corresponding to each matrix (data size: nm × 1) was used as the output of the CNN3 model. The number of iterations was 100 to complete the training of the CNN3 model parameters (model structure and parameters are shown in Table 2). ④ The n × m matrix in the validation set was used as the input of the CNN3 model, and the model output was compared with the damage sub-region corresponding to the validation set matrix (ideal output) to verify the model parameters.

[0074] Table 2 Secondary CNN3 model parameters

[0075]

[0076]

[0077] In practical applications, the normalized damage scattering signal envelope collected by the sensor array is input into the trained two-stage CNN network prediction model to achieve rapid and accurate identification of the damaged area of the carbon fiber composite material structure.

[0078] Example 2

[0079] This embodiment provides a two-stage CNN model training system.

[0080] A two-stage CNN model training system, comprising:

[0081] A database construction module is configured to: construct a measured simulated damage sample data set and a numerically calculated damage sample data set; and construct a structural damage sample database based on the measured simulated damage sample data set and the numerically calculated damage sample data set;

[0082] The model training module is configured to: use a structural damage sample database to train the first CNN model and the second CNN model respectively, to obtain the first feature matrix and the second feature matrix respectively; and train the third CNN model based on the fusion matrix of the first feature matrix and the second feature matrix.

[0083] It should be noted that the examples and application scenarios implemented by the above-mentioned database construction module and model training module are the same as those in the steps of Example 1, but are not limited to the contents disclosed in the above-mentioned Example 1. It should be noted that the above-mentioned modules, as part of the system, can be executed in a computer system such as a set of computer-executable instructions.

[0084] Example 3

[0085] This embodiment provides a two-stage CNN model training system.

[0086] A two-stage CNN model training system includes: a piezoelectric sensor array, an ultrasonic guided wave monitoring instrument, and a host computer, wherein the piezoelectric sensor array is coupled to a carbon fiber composite material structure, the piezoelectric sensor array is connected to the ultrasonic guided wave monitoring instrument, and the ultrasonic guided wave monitoring instrument is connected to the host computer; wherein,

[0087] The piezoelectric sensor array is used to collect guided wave signals when the structure is in a healthy state and guided wave signals after the structure is damaged, and transmit them to the host computer through the ultrasonic guided wave monitoring instrument;

[0088] The host computer obtains a scattered signal envelope based on the guided wave signal in a healthy state of the structure and the guided wave signal after the structure is damaged, and constructs a measured simulated damage sample data set of the scattered signal envelope; and fits the propagation speed of Lamb waves in various directions of the carbon fiber composite material structure based on the guided wave signal in a healthy state of the structure to construct a numerical calculation damage sample data set; constructs a structural damage sample database based on the measured simulated damage sample data set and the numerical calculation damage sample data set; uses the structural damage sample database to train a first CNN model and a second CNN model, respectively, to obtain a first feature matrix and a second feature matrix, respectively; and trains a third CNN model based on a fusion matrix of the first feature matrix and the second feature matrix.

[0089] The specific solution of this embodiment is as follows:

[0090] The two-stage CNN model training system mainly includes a carbon fiber composite structure, a piezoelectric transducer (PZT) array, an ultrasonic guided wave monitoring instrument, and a host computer. ① One PZT sensor is attached to the surface of the composite structure to form a square sensor array containing L (L = l·(l-1) / 2) sensing paths, which is responsible for the excitation and reception of Lamb waves. ② The ultrasonic guided wave monitoring instrument is connected to the sensor array at one end and to the host computer at the other end. It has functions such as channel switching, energy amplification, and high-speed data acquisition, and is mainly responsible for the transmission of excitation and reception signals between the sensor array and the host computer. ③ The host computer software is connected to the ultrasonic guided wave monitoring instrument and has functions such as Lamb wave generation, instrument parameter setting, waveform display, and storage. It is mainly responsible for the generation of excitation guided waves and the storage of received guided waves.

[0091] A "one-transmit-multiple-receive" "excitation-receive" patrol mode (the excitation signal is a five-peak Lamb wave with a center frequency of 50kHz modulated by a Hanning window) is adopted. Damage is simulated with absorbing mud to establish a measured simulated damage sample data set: ① A carbon fiber composite material structure damage identification system is established. The guided wave signal (signal duration is T) in the healthy state of the structure is collected based on the PZT sensor array and marked as the healthy signal; ② The absorbing mud is placed at the pre-demarcated simulated damage position, and the guided wave signal after the structure is damaged is collected based on the PZT sensor array and marked as the damaged signal; ③ The scattering signal corresponding to the damage is calculated (scattering signal = damage signal - healthy signal); ④ The envelope of the scattering signal is calculated using Hilbert transform and normalized. The amplitude of the normalized scattering signal envelope is within [0,1]; ⑤ The set of normalized scattering signal envelopes of all simulated damage discrete points in the monitoring area is used as the measured simulated damage sample data set.

[0092] like Figure 1(a)-Figure 1(e) As shown in the figure, based on the guided wave signal in the healthy state of the structure, the group propagation velocity of the S0 mode wave in any direction of the structure is fitted: ① Define the signal propagation time t on the i-th path i is the peak moment of S0 direct wave envelope ri and the peak moment of the excitation signal envelope t ai The time difference, that is: t i =t ai -t ri ;② Combined with the length d of the i-th sensing path i , calculate the group propagation velocity v of the S0 mode wave in the direction of the i-th path i =d i / t i; ③ Calculate the average value of the propagation velocity with the same propagation angle (θ±90°), and then fit the average propagation velocity in the range of 0° to 90° with a quadratic polynomial to obtain v=f(θ)=a·θ 2 +b·θ+c (a, b and c are fitting parameters); ④ Based on the velocity fitting results in the range of 0° to 90°, the mapping function v(θ) between the group propagation velocity v and the propagation angle θ of the S0 mode with a center frequency of 50 kHz in all directions of the carbon fiber composite material structure is obtained, as shown in formula (1).

[0093]

[0094] Based on the pre-fitted propagation velocity of Lamb waves in all directions of carbon fiber composite materials, a numerical calculation damage sample data set is established: ① According to the damage identification accuracy requirement, the entire monitoring area is divided into n×m sub-monitoring areas (area 1 to area nm); ② For any position (x, y) within sub-area k (1≤k≤nm), if damage occurs here, the arrival time of the scattered signal corresponding to the damage detected by the sensing path i can be expressed by formula (2); ③ When damage occurs at a discrete point, ideally, the amplitude-normalized S0 mode wave received on the sensing path i directly reaches the scattered signal (scattered signal = damage signal - healthy signal) and its Hilbert envelope can be time-shifted backward by the excitation signal and its Hilbert envelope At this moment, the normalized scattered signal duration = the excitation signal duration, the normalized scattered signal envelope peak = the normalized excitation signal envelope peak, ④ Taking into account the size of the sub-monitoring area and the damage identification accuracy, each sub-monitoring area is divided into multiple discrete damage points. Then, the normalized scattered signal envelope at each point is calculated according to steps ②-③. ⑤ The set of normalized scattered signal envelopes of all numerically calculated damage discrete points in the monitoring area is used as the numerically calculated damage sample data set.

[0095]

[0096] In formula (2), (x ia ,y ia ) and (x ir ,y ir ) represent the position coordinates of the exciter and receiver on sensing path i, and Represent the distances from the exciter and receiver to the discrete points, and represent the group propagation velocity of the S0 mode wave in the directions of the exciter-discrete point and the receiver-discrete point, respectively. and represent the propagation angles of the exciter-discrete point and receiver-discrete point directions relative to the sensing path i, and Represent the time for the S0 mode wave to propagate from the exciter to the discrete point and from the discrete point to the receiver, respectively. is the total time taken for the S0 mode wave to propagate from the exciter to the discrete point and then to the receiver.

[0097] Forming a carbon fiber composite material structure damage sample database: All normalized scattering signal Hilbert envelopes contained in the measured simulated damage sample data set and the numerical calculation damage sample data set are used together as the carbon fiber composite material structure damage sample database.

[0098] This example uses numerical calculation data and a small amount of measured data to construct a database of guided wave signals for structural damage. Three sets of convolutional neural networks (CNNs) with different sizes and structures are then combined to establish a model for identifying damaged areas in carbon fiber composite materials using a two-stage CNN network and ultrasonic guided waves. The primary CNN model set is used for preliminary damage area identification, while the secondary CNN model is used for final determination of the damage area.

[0099] The specific solution of this embodiment can be implemented by referring to the following content:

[0100] First, a numerical damage sample dataset was established based on the fitted Lamb wave propagation velocities in various directions within the carbon fiber structure and the coordinates of each discrete point within the monitoring area. Combined with the measured simulated damage sample dataset, a damage sample database within the structural monitoring area was constructed for damage prediction model training. Then, using the normalized scattered signal envelope from the damage sample database as input and the damaged area as output, two sets of primary damage region recognition models (CNN1 and CNN2) based on CNN networks with different region recognition sizes were trained. Finally, the preliminary damage region recognition results obtained by CNN1 and CNN2 were used as input features, and the damaged area was used as the target output to establish a secondary damage region recognition model (CNN3).

[0101] This embodiment combines the theoretically calculated guided wave scattering signal envelope with the actual collected scattering signal envelope of the absorbing mud simulated damage to quickly and cheaply construct any number of carbon fiber composite material structure damage data sample libraries, effectively solving the problem that measured damage data is difficult to obtain in large quantities, but deep neural network models require massive sample data to drive them.

[0102] Example 4

[0103] This embodiment provides a damaged area identification method.

[0104] like Figure 4 As shown, a damaged area identification method includes:

[0105] Based on the scattered signal of the damage to be detected, the signals are input into the first CNN model and the second CNN model respectively to obtain a first prediction result matrix and a second prediction result matrix respectively;

[0106] Obtaining a preliminary prediction result matrix based on the product of the first prediction result matrix and the second prediction result matrix;

[0107] The third CNN model is used to predict the preliminary prediction result matrix to obtain the regional prediction result of the damage to be tested;

[0108] Among them, the first CNN model, the second CNN model and the third CNN model are all obtained through the two-stage CNN model training method described in Example 1.

[0109] Based on the established two-stage damage region recognition model, regional recognition of the damage to be tested is performed: ① The normalized envelope of the scattered signal of the damage to be tested collected by the sensor array (data size: T×L) is input into the trained CNN model groups CNN1 and CNN2 respectively. Then, based on the prediction results output by the model groups, an n×m-dimensional preliminary prediction result statistical matrix is constructed, where the element at position (g, j) in the matrix is the product of the number of occurrences of the region (g-1)×m+j in the CNN1 prediction result and the CNN2 prediction result. ② The preliminary prediction result statistical matrix n×m is input into the trained CNN3 model, and the model output is the regional prediction result of the damage to be tested.

[0110] This embodiment establishes a two-level damage area identification model by using CNN networks of different sizes and structures. It can not only effectively reduce the extremely high requirements for the regional resolution ability of the deep learning network model in large-area monitoring areas, but also reduce the excessive dependence of the model performance on parameters, thereby improving the accuracy of damage area identification in carbon fiber composite materials.

[0111] Example 5

[0112] This embodiment provides a damaged area identification system.

[0113] A damaged area identification system, comprising:

[0114] A primary prediction module is configured to: input the scattered signals of the damage to be detected into the first CNN model and the second CNN model respectively, to obtain a first prediction result matrix and a second prediction result matrix respectively;

[0115] A fusion module is configured to: obtain a preliminary prediction result matrix based on the product of the first prediction result matrix and the second prediction result matrix;

[0116] The secondary prediction module is configured to: use the third CNN model to predict the preliminary prediction result matrix to obtain the regional prediction result of the damage to be detected;

[0117] Among them, the first CNN model, the second CNN model and the third CNN model are all obtained through the two-stage CNN model training method described in the first aspect.

[0118] It should be noted that the examples and application scenarios implemented by the above-mentioned primary prediction module, fusion module, and secondary prediction module are the same as those in the steps of Example 4, but are not limited to the contents disclosed in the above-mentioned Example 4. It should be noted that the above-mentioned modules, as part of the system, can be executed in a computer system such as a set of computer-executable instructions.

[0119] Example 6

[0120] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the program implements the steps in the two-stage CNN model training method described in Example 1, or implements the steps in the damage area identification method described in Example 4.

[0121] Example 7

[0122] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the two-stage CNN model training method described in Example 1 are implemented, or the steps of the damage area identification method described in Example 4 are implemented.

[0123] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage and optical storage, etc.) containing computer-usable program code.

[0124] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0125] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0126] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0127] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0128] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A two-stage CNN model training method for structural damage area identification, characterized by: include: Constructing a dataset of measured and simulated damage samples and a dataset of numerically calculated damage samples; Based on the measured and simulated damage sample data set and the numerical calculation damage sample data set, a structural damage sample database is constructed; Using the structural damage sample database, the primary damage area recognition model set CNN1 and the primary damage area recognition model set CNN2 are trained respectively to obtain the first feature matrix and the second feature matrix respectively. Based on the fusion matrix of the first feature matrix and the second feature matrix, the third CNN model is trained, specifically: Based on the CNN network, a primary damage area recognition model set CNN1 containing (n-1)×(m-1) CNN models is established: (1) All sub-monitoring areas are divided into (n-1)×(m-1) different area sets in order according to the size of 2×2 and the step size of 1; (2) The damage data samples in each area set are randomly generated into training sets and validation sets in a ratio of 8:2; (3) The normalized scattered signal envelope in the training set is used as the input of the CNN network model, and the sub-area corresponding to the normalized envelope is used as the output of the CNN network model. The number of iterations is 100 to complete the network model parameter training; (4) The normalized scattered signal envelope in the validation set is used as the model input, and the model output is compared with the data label in the validation set to verify the model parameters; (5) Repeat steps (3)-(4) for all area sets to obtain the primary damage area recognition model set CNN1; The same steps as steps (1)-(5) are used to establish the primary damage area recognition model set CNN2; The normalized envelope of the scattered signal of the damage to be measured collected by the sensor array is input into the trained CNN model groups CNN1 and CNN2 respectively. Then, based on the prediction results output by the model groups, an n×m-dimensional preliminary prediction result statistical matrix is constructed, where the element at position (g, j) in the matrix is the product of the number of occurrences of the region (g-1)×m+j in the CNN1 prediction result and the CNN2 prediction result. The preliminary prediction result statistical matrix n×m is input into the trained CNN3 model, and the model output result is the regional prediction result of the damage to be measured.

2. The two-stage CNN model training method for structural damage area identification according to claim 1 is characterized in that: The process of constructing the measured simulated damage sample data set includes: Obtain the guided wave signal when the structure is in a healthy state and mark it as a healthy signal; Simulate structural damage and obtain the guided wave signal after the structure is damaged, which is marked as the damage signal; According to the healthy signal and the damaged signal, the scattered signal corresponding to the damaged signal is obtained; The envelope of the scattered signal is calculated using Hilbert transform, and a dataset of measured simulated damage samples of the scattered signal envelope is constructed.

3. The two-stage CNN model training method for structural damage area identification according to claim 1, characterized in that: The process of constructing the numerical calculation damage sample data set includes: Based on the guided wave signal in the healthy state of the structure, the propagation velocity of Lamb waves in each direction of the carbon fiber composite material structure is fitted; Based on the pre-fitted propagation velocity of Lamb waves in all directions of the carbon fiber composite material structure, a numerical calculation damage sample data set is constructed.

4. A two-stage CNN model training system for structural damage area identification, characterized by: include: A database construction module is configured to: construct a measured simulated damage sample data set and a numerically calculated damage sample data set; Based on the measured and simulated damage sample data set and the numerical calculation damage sample data set, a structural damage sample database is constructed; The model training module is configured to: use the structural damage sample database to train the primary damage area identification model set CNN1 and the primary damage area identification model set CNN2, respectively, to obtain the first feature matrix and the second feature matrix; based on the fusion matrix of the first feature matrix and the second feature matrix, train the third CNN model, specifically: Based on the CNN network, a primary damage area recognition model set CNN1 containing (n-1)×(m-1) CNN models is established: (1) All sub-monitoring areas are divided into (n-1)×(m-1) different area sets in order according to the size of 2×2 and the step size of 1; (2) The damage data samples in each area set are randomly generated into training sets and validation sets in a ratio of 8:2; (3) The normalized scattered signal envelope in the training set is used as the input of the CNN network model, and the sub-area corresponding to the normalized envelope is used as the output of the CNN network model. The number of iterations is 100 to complete the network model parameter training; (4) The normalized scattered signal envelope in the validation set is used as the model input, and the model output is compared with the data label in the validation set to verify the model parameters; (5) Repeat steps (3)-(4) for all area sets to obtain the primary damage area recognition model set CNN1; The same steps as steps (1)-(5) are used to establish the primary damage area recognition model set CNN2; The normalized envelope of the scattered signal of the damage to be measured collected by the sensor array is input into the trained CNN model groups CNN1 and CNN2 respectively. Then, based on the prediction results output by the model groups, an n×m-dimensional preliminary prediction result statistical matrix is constructed, where the element at position (g, j) in the matrix is the product of the number of occurrences of the region (g-1)×m+j in the CNN1 prediction result and the CNN2 prediction result. The preliminary prediction result statistical matrix n×m is input into the trained CNN3 model, and the model output result is the regional prediction result of the damage to be measured.

5. A two-stage CNN model training system for structural damage area identification, characterized by: include: A piezoelectric sensor array, an ultrasonic guided wave monitoring instrument, and a host computer, wherein the piezoelectric sensor array is coupled to a carbon fiber composite material structure, the piezoelectric sensor array is connected to the ultrasonic guided wave monitoring instrument, and the ultrasonic guided wave monitoring instrument is connected to the host computer; wherein, The piezoelectric sensor array is used to collect guided wave signals when the structure is in a healthy state and guided wave signals after the structure is damaged, and transmit them to the host computer through the ultrasonic guided wave monitoring instrument; The host computer obtains a scattered signal envelope based on the guided wave signal in a healthy state of the structure and the guided wave signal after the structure is damaged, and constructs a measured simulated damage sample data set of the scattered signal envelope; and fits the propagation velocity of Lamb waves in various directions of the carbon fiber composite material structure based on the guided wave signal in a healthy state of the structure, thereby constructing a numerical calculation damage sample data set; constructs a structural damage sample database based on the measured simulated damage sample data set and the numerical calculation damage sample data set; uses the structural damage sample database to train a primary damage area identification model set CNN1 and a primary damage area identification model set CNN2, respectively, to obtain a first feature matrix and a second feature matrix, respectively; and trains a third CNN model based on a fusion matrix of the first feature matrix and the second feature matrix, specifically: Based on the CNN network, a primary damage area recognition model set CNN1 containing (n-1)×(m-1) CNN models is established: (1) All sub-monitoring areas are divided into (n-1)×(m-1) different area sets in order according to the size of 2×2 and the step size of 1; (2) The damage data samples in each area set are randomly generated into training sets and validation sets in a ratio of 8:2; (3) The normalized scattered signal envelope in the training set is used as the input of the CNN network model, and the sub-area corresponding to the normalized envelope is used as the output of the CNN network model. The number of iterations is 100 to complete the network model parameter training; (4) The normalized scattered signal envelope in the validation set is used as the model input, and the model output is compared with the data label in the validation set to verify the model parameters; (5) Repeat steps (3)-(4) for all area sets to obtain the primary damage area recognition model set CNN1; The same steps as steps (1)-(5) are used to establish the primary damage area recognition model set CNN2; The normalized envelope of the scattered signal of the damage to be measured collected by the sensor array is input into the trained CNN model groups CNN1 and CNN2 respectively. Then, based on the prediction results output by the model groups, an n×m-dimensional preliminary prediction result statistical matrix is constructed, where the element at position (g, j) in the matrix is the product of the number of occurrences of the region (g-1)×m+j in the CNN1 prediction result and the CNN2 prediction result. The preliminary prediction result statistical matrix n×m is input into the trained CNN3 model, and the model output result is the regional prediction result of the damage to be measured.

6. A method for identifying a damaged area, characterized in that: include: Based on the scattered signal of the damage to be detected, the signals are input into the first CNN model and the second CNN model respectively to obtain a first prediction result matrix and a second prediction result matrix respectively; Obtaining a preliminary prediction result matrix based on the product of the first prediction result matrix and the second prediction result matrix; The third CNN model is used to predict the preliminary prediction result matrix to obtain the regional prediction result of the damage to be tested; The first CNN model, the second CNN model and the third CNN model are all obtained by the two-stage CNN model training method described in any one of claims 1 to 3.

7. The damaged area identification method according to claim 6, characterized in that: The preliminary prediction result matrix is obtained, including: If the preliminary prediction result matrix is an n × m dimension preliminary prediction result matrix; Then the position in the matrix ( g , j ) is the area between the first CNN model prediction result and the second CNN model prediction result (( g - 1) × m + j ) multiplied by the number of occurrences.

8. The damage area identification system is characterized by: include: A primary prediction module is configured to: input the scattered signals of the damage to be detected into the first CNN model and the second CNN model respectively, to obtain a first prediction result matrix and a second prediction result matrix respectively; A fusion module is configured to: obtain a preliminary prediction result matrix based on the product of the first prediction result matrix and the second prediction result matrix; The secondary prediction module is configured to: use the third CNN model to predict the preliminary prediction result matrix to obtain the regional prediction result of the damage to be detected; The first CNN model, the second CNN model and the third CNN model are all obtained by the two-stage CNN model training method described in any one of claims 1 to 3.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, it implements the steps in the two-stage CNN model training method as described in any one of claims 1-3, or implements the steps in the damage area identification method as described in any one of claims 6-7.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, it implements the steps in the two-stage CNN model training method as described in any one of claims 1-3, or implements the steps in the damage area identification method as described in any one of claims 6-7.

Images